Gemini 4 Argon launch notes: pricing, caching and agent features
A thread rounds up Argon's launch notes, pricing and model card: $2/$10 per million tokens until November 30, then $4/$20, a 95% discount on cached input, and a focus on long agent runs and prompt-injection resistance.
Google just published everything about Gemini 4 Argon: the launch notes, the pricing and the model card read this and you'll use Argon better than 95% of people what it is what it costs how to use it before the price doubles > step 1 - it's the first of a family Argon is Google's new frontier model, and the first release of Gemini 4 more models are coming after it > step 2 - the price window $2 in, $10 out per 1M tokens until November 30 then $4 / $20, the same list price as Claude Opus 5.5 > step 3 - cache everything cached input is 95% off a stable system prompt makes most of your input nearly free > step 4 - one budget for everything 1M tokens shared between input and output long outputs keep going past the cap with a continuation protocol > step 5 - where to get it API and AI Studio, live globally today Gemini app and AI Mode for Pro and Ultra, Antigravity for Ultra and enterprise > step 6 - built for long agent runs multi-agent loops, /goal commands and custom skills inside Antigravity > step 7 - coding inside Google, agentic code submissions went up 35% and code rollbacks dropped by nearly a third > step 8 - it writes like you asked follows style instructions better, hedges less, repeats itself less > step 9 - harder to hijack prompt injection succeeded 0.7% of the time across 15 adaptive attempts shortcut attempts on coding benchmarks dropped to 2% > step 10 - the numbers are Google's #1 on Vals Index, 87% on ERQA run 20 of your own tasks on it before you switch anything the result: frontier tokens at half price for 6 more weeks then the same price as Opus, and your own tests decide
This post quotes another post: x.com/mahaximus_/status/2085024744387092973